Standard mode

Standard mode is the plain OpenAI Chat Completions format. If your app already talks to an OpenAI-compatible API, change two things and it works:

Setting	Value
Base URL	https://api.infernear.dev/v1
API key	your inf_... key

The model still runs inside NEAR AI's attested TEE. What standard mode does not give you is privacy from InferNear's relay: the relay can read the prompt and the reply while it passes them along. It stores neither. If that matters, use private mode.

Examples

Python

# pip install openai
import os
from openai import OpenAI

client = OpenAI(base_url='https://api.infernear.dev/v1', api_key=os.environ['INFERNEAR_API_KEY'])
result = client.chat.completions.create(
    model='Qwen/Qwen3-VL-30B-A3B-Instruct',
    messages=[{'role': 'user', 'content': 'Reply with the word ok.'}],
    max_tokens=64,
)
print(result.choices[0].message.content)

JavaScript

// npm install openai
import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://api.infernear.dev/v1',
  apiKey: process.env.INFERNEAR_API_KEY,
});
const result = await client.chat.completions.create({
  model: 'Qwen/Qwen3-VL-30B-A3B-Instruct',
  messages: [{ role: 'user', content: 'Reply with the word ok.' }],
  max_tokens: 64,
});
console.log(result.choices[0].message.content);

curl

curl -s https://api.infernear.dev/v1/chat/completions \
  -H "Authorization: Bearer $INFERNEAR_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"model":"Qwen/Qwen3-VL-30B-A3B-Instruct","messages":[{"role":"user","content":"Reply with the single word: ok"}],"max_tokens":64}'

What the relay changes in your request

Billing needs one explicit reply limit and the token usage, so the relay normalizes two things. Nothing else in the body is changed.

Field	What the relay does
max_tokens	If absent, it is set from max_completion_tokens, or else to 4,096
max_completion_tokens	Removed: the request goes upstream with max_tokens only
stream_options.include_usage	Set to true for streaming requests, even if the request said false

What is supported

• Streaming ("stream": true) and non-streaming replies. Response bytes, including the stream framing, are passed through unchanged.
• Tool definitions and tool calls.
• Text input and output.

What is refused

Request	Result
max_tokens above 4,096	400
n other than 1	400
Input that is not text	400
A model id that is not in GET /v1/models (including aliases)	404
A body larger than 1 MiB	413
More than 4 requests in flight for one user	429

Tips

• Give reasoning models room. With a small max_tokens (such as 64), a reasoning model can spend the whole budget thinking and return no visible answer. The call is still charged for the tokens used.
• A retry is a new call. Client idempotency keys are not implemented, so a retried request is a new inference with its own charge.
• Check the billing header. X-InferNear-Billing tells you whether a call is billed as free or to credits.

Full details of routes, headers and errors: API reference.